Optimisation, games and learning strategies in telecommunication systems subject to structural constraints

Massaro, Antonio (2019) Optimisation, games and learning strategies in telecommunication systems subject to structural constraints. PhD thesis, University of Trento.

[img]PDF - Doctoral Thesis
Restricted to Repository staff only until 9999.
Available under License Creative Commons Attribution.

[img]PDF - Disclaimer
Restricted to Repository staff only until 9999.



Telecommunication systems are becoming more and more complex and dynamic. As an example, on e-commerce platforms, on content provision platforms or cloud computing platforms, a variety of dierent agents interact and inuence each other. In data-centers, large numbers of dierent ows need to be served meeting a given quality of service, and the routing decisions taken for one ow can easily inuence the performance of some other ows. These examples are characterized by complexity and dynamism. These features emerge form the interactions among agents and the environment where they operate, and from the fact that agents can choose their actions within a spectrum of possible strategies. As a result, given the structural constraints of a system, it is usually challenging to understand what the emerging properties and the operating points of such system are. In the present work we analyze the dynamics of such systems, with the aim of devising strategies to optimize their operating points. Given their variety, we resort to dierent techniques to analyze each of them. In the second and third chapters we will resort to a game theoretical framework, as we will need to model an edge caching system as a multi-agent system where the decisions made by one of the agents determine the behavior of all the others. In the fourth chapter we will tackle a ow segmentation problem. In this context we will need to model the behavior of an agent seeking to maximize a private utility from the interactions with a stochastic environment, therefore we will use the theory of stochastic optimization. In the fth chapter we tackle the problem of determining an optimal trunk-reservation policy for a queue control problem, in absence of knowledge of the statistical properties of the queue. In this case, we will use reinforcement learning tools to devise an algorithm based on policy learning that converges to an optimal policy.

Item Type:Doctoral Thesis (PhD)
Doctoral School:Information and Communication Technology
PhD Cycle:31
Subjects:Area 01 - Scienze matematiche e informatiche > INF/01 INFORMATICA
Repository Staff approval on:21 May 2019 10:13

Repository Staff Only: item control page